2 * Copyright (C) 2012,2013 - ARM Ltd
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program. If not, see <http://www.gnu.org/licenses/>.
18 #ifndef __ARM64_KVM_MMU_H__
19 #define __ARM64_KVM_MMU_H__
22 #include <asm/memory.h>
23 #include <asm/cpufeature.h>
26 * As ARMv8.0 only has the TTBR0_EL2 register, we cannot express
27 * "negative" addresses. This makes it impossible to directly share
28 * mappings with the kernel.
30 * Instead, give the HYP mode its own VA region at a fixed offset from
31 * the kernel by just masking the top bits (which are all ones for a
32 * kernel address). We need to find out how many bits to mask.
34 * We want to build a set of page tables that cover both parts of the
35 * idmap (the trampoline page used to initialize EL2), and our normal
36 * runtime VA space, at the same time.
38 * Given that the kernel uses VA_BITS for its entire address space,
39 * and that half of that space (VA_BITS - 1) is used for the linear
40 * mapping, we can also limit the EL2 space to (VA_BITS - 1).
42 * The main question is "Within the VA_BITS space, does EL2 use the
43 * top or the bottom half of that space to shadow the kernel's linear
44 * mapping?". As we need to idmap the trampoline page, this is
45 * determined by the range in which this page lives.
47 * If the page is in the bottom half, we have to use the top half. If
48 * the page is in the top half, we have to use the bottom half:
50 * T = __pa_symbol(__hyp_idmap_text_start)
51 * if (T & BIT(VA_BITS - 1))
52 * HYP_VA_MIN = 0 //idmap in upper half
54 * HYP_VA_MIN = 1 << (VA_BITS - 1)
55 * HYP_VA_MAX = HYP_VA_MIN + (1 << (VA_BITS - 1)) - 1
57 * This of course assumes that the trampoline page exists within the
58 * VA_BITS range. If it doesn't, then it means we're in the odd case
59 * where the kernel idmap (as well as HYP) uses more levels than the
60 * kernel runtime page tables (as seen when the kernel is configured
61 * for 4k pages, 39bits VA, and yet memory lives just above that
62 * limit, forcing the idmap to use 4 levels of page tables while the
63 * kernel itself only uses 3). In this particular case, it doesn't
64 * matter which side of VA_BITS we use, as we're guaranteed not to
65 * conflict with anything.
67 * When using VHE, there are no separate hyp mappings and all KVM
68 * functionality is already mapped as part of the main kernel
69 * mappings, and none of this applies in that case.
72 #define HYP_PAGE_OFFSET_HIGH_MASK ((UL(1) << VA_BITS) - 1)
73 #define HYP_PAGE_OFFSET_LOW_MASK ((UL(1) << (VA_BITS - 1)) - 1)
77 #include <asm/alternative.h>
78 #include <asm/cpufeature.h>
81 * Convert a kernel VA into a HYP VA.
82 * reg: VA to be converted.
84 * This generates the following sequences:
86 * and x0, x0, #HYP_PAGE_OFFSET_HIGH_MASK
89 * and x0, x0, #HYP_PAGE_OFFSET_HIGH_MASK
90 * and x0, x0, #HYP_PAGE_OFFSET_LOW_MASK
95 * The "low mask" version works because the mask is a strict subset of
96 * the "high mask", hence performing the first mask for nothing.
97 * Should be completely invisible on any viable CPU.
99 .macro kern_hyp_va reg
100 alternative_if_not ARM64_HAS_VIRT_HOST_EXTN
101 and \reg, \reg, #HYP_PAGE_OFFSET_HIGH_MASK
102 alternative_else_nop_endif
103 alternative_if ARM64_HYP_OFFSET_LOW
104 and \reg, \reg, #HYP_PAGE_OFFSET_LOW_MASK
105 alternative_else_nop_endif
110 #include <asm/pgalloc.h>
111 #include <asm/cache.h>
112 #include <asm/cacheflush.h>
113 #include <asm/mmu_context.h>
114 #include <asm/pgtable.h>
116 static inline unsigned long __kern_hyp_va(unsigned long v)
118 asm volatile(ALTERNATIVE("and %0, %0, %1",
120 ARM64_HAS_VIRT_HOST_EXTN)
122 : "i" (HYP_PAGE_OFFSET_HIGH_MASK));
123 asm volatile(ALTERNATIVE("nop",
125 ARM64_HYP_OFFSET_LOW)
127 : "i" (HYP_PAGE_OFFSET_LOW_MASK));
131 #define kern_hyp_va(v) ((typeof(v))(__kern_hyp_va((unsigned long)(v))))
134 * Obtain the PC-relative address of a kernel symbol
137 * The goal of this macro is to return a symbol's address based on a
138 * PC-relative computation, as opposed to a loading the VA from a
139 * constant pool or something similar. This works well for HYP, as an
140 * absolute VA is guaranteed to be wrong. Only use this if trying to
141 * obtain the address of a symbol (i.e. not something you obtained by
142 * following a pointer).
144 #define hyp_symbol_addr(s) \
147 asm("adrp %0, %1\n" \
148 "add %0, %0, :lo12:%1\n" \
149 : "=r" (addr) : "S" (&s)); \
154 * We currently only support a 40bit IPA.
156 #define KVM_PHYS_SHIFT (40)
157 #define KVM_PHYS_SIZE (1UL << KVM_PHYS_SHIFT)
158 #define KVM_PHYS_MASK (KVM_PHYS_SIZE - 1UL)
160 #include <asm/stage2_pgtable.h>
162 int create_hyp_mappings(void *from, void *to, pgprot_t prot);
163 int create_hyp_io_mappings(void *from, void *to, phys_addr_t);
164 void free_hyp_pgds(void);
166 void stage2_unmap_vm(struct kvm *kvm);
167 int kvm_alloc_stage2_pgd(struct kvm *kvm);
168 void kvm_free_stage2_pgd(struct kvm *kvm);
169 int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
170 phys_addr_t pa, unsigned long size, bool writable);
172 int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run);
174 void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu);
176 phys_addr_t kvm_mmu_get_httbr(void);
177 phys_addr_t kvm_get_idmap_vector(void);
178 int kvm_mmu_init(void);
179 void kvm_clear_hyp_idmap(void);
181 #define kvm_set_pte(ptep, pte) set_pte(ptep, pte)
182 #define kvm_set_pmd(pmdp, pmd) set_pmd(pmdp, pmd)
184 static inline pte_t kvm_s2pte_mkwrite(pte_t pte)
186 pte_val(pte) |= PTE_S2_RDWR;
190 static inline pmd_t kvm_s2pmd_mkwrite(pmd_t pmd)
192 pmd_val(pmd) |= PMD_S2_RDWR;
196 static inline void kvm_set_s2pte_readonly(pte_t *pte)
198 pteval_t old_pteval, pteval;
200 pteval = READ_ONCE(pte_val(*pte));
203 pteval &= ~PTE_S2_RDWR;
204 pteval |= PTE_S2_RDONLY;
205 pteval = cmpxchg_relaxed(&pte_val(*pte), old_pteval, pteval);
206 } while (pteval != old_pteval);
209 static inline bool kvm_s2pte_readonly(pte_t *pte)
211 return (pte_val(*pte) & PTE_S2_RDWR) == PTE_S2_RDONLY;
214 static inline void kvm_set_s2pmd_readonly(pmd_t *pmd)
216 kvm_set_s2pte_readonly((pte_t *)pmd);
219 static inline bool kvm_s2pmd_readonly(pmd_t *pmd)
221 return kvm_s2pte_readonly((pte_t *)pmd);
224 static inline bool kvm_page_empty(void *ptr)
226 struct page *ptr_page = virt_to_page(ptr);
227 return page_count(ptr_page) == 1;
230 #define hyp_pte_table_empty(ptep) kvm_page_empty(ptep)
232 #ifdef __PAGETABLE_PMD_FOLDED
233 #define hyp_pmd_table_empty(pmdp) (0)
235 #define hyp_pmd_table_empty(pmdp) kvm_page_empty(pmdp)
238 #ifdef __PAGETABLE_PUD_FOLDED
239 #define hyp_pud_table_empty(pudp) (0)
241 #define hyp_pud_table_empty(pudp) kvm_page_empty(pudp)
246 #define kvm_flush_dcache_to_poc(a,l) __flush_dcache_area((a), (l))
248 static inline bool vcpu_has_cache_enabled(struct kvm_vcpu *vcpu)
250 return (vcpu_sys_reg(vcpu, SCTLR_EL1) & 0b101) == 0b101;
253 static inline void __coherent_cache_guest_page(struct kvm_vcpu *vcpu,
257 void *va = page_address(pfn_to_page(pfn));
259 kvm_flush_dcache_to_poc(va, size);
261 if (icache_is_aliasing()) {
262 /* any kind of VIPT cache */
263 __flush_icache_all();
264 } else if (is_kernel_in_hyp_mode() || !icache_is_vpipt()) {
265 /* PIPT or VPIPT at EL2 (see comment in __kvm_tlb_flush_vmid_ipa) */
266 flush_icache_range((unsigned long)va,
267 (unsigned long)va + size);
271 static inline void __kvm_flush_dcache_pte(pte_t pte)
273 struct page *page = pte_page(pte);
274 kvm_flush_dcache_to_poc(page_address(page), PAGE_SIZE);
277 static inline void __kvm_flush_dcache_pmd(pmd_t pmd)
279 struct page *page = pmd_page(pmd);
280 kvm_flush_dcache_to_poc(page_address(page), PMD_SIZE);
283 static inline void __kvm_flush_dcache_pud(pud_t pud)
285 struct page *page = pud_page(pud);
286 kvm_flush_dcache_to_poc(page_address(page), PUD_SIZE);
289 #define kvm_virt_to_phys(x) __pa_symbol(x)
291 void kvm_set_way_flush(struct kvm_vcpu *vcpu);
292 void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled);
294 static inline bool __kvm_cpu_uses_extended_idmap(void)
296 return __cpu_uses_extended_idmap();
300 * Can't use pgd_populate here, because the extended idmap adds an extra level
301 * above CONFIG_PGTABLE_LEVELS (which is 2 or 3 if we're using the extended
302 * idmap), and pgd_populate is only available if CONFIG_PGTABLE_LEVELS = 4.
304 static inline void __kvm_extend_hypmap(pgd_t *boot_hyp_pgd,
306 pgd_t *merged_hyp_pgd,
307 unsigned long hyp_idmap_start)
312 * Use the first entry to access the HYP mappings. It is
313 * guaranteed to be free, otherwise we wouldn't use an
316 VM_BUG_ON(pgd_val(merged_hyp_pgd[0]));
317 merged_hyp_pgd[0] = __pgd(__pa(hyp_pgd) | PMD_TYPE_TABLE);
320 * Create another extended level entry that points to the boot HYP map,
321 * which contains an ID mapping of the HYP init code. We essentially
322 * merge the boot and runtime HYP maps by doing so, but they don't
323 * overlap anyway, so this is fine.
325 idmap_idx = hyp_idmap_start >> VA_BITS;
326 VM_BUG_ON(pgd_val(merged_hyp_pgd[idmap_idx]));
327 merged_hyp_pgd[idmap_idx] = __pgd(__pa(boot_hyp_pgd) | PMD_TYPE_TABLE);
330 static inline unsigned int kvm_get_vmid_bits(void)
332 int reg = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
334 return (cpuid_feature_extract_unsigned_field(reg, ID_AA64MMFR1_VMIDBITS_SHIFT) == 2) ? 16 : 8;
338 * We are not in the kvm->srcu critical section most of the time, so we take
339 * the SRCU read lock here. Since we copy the data from the user page, we
340 * can immediately drop the lock again.
342 static inline int kvm_read_guest_lock(struct kvm *kvm,
343 gpa_t gpa, void *data, unsigned long len)
345 int srcu_idx = srcu_read_lock(&kvm->srcu);
346 int ret = kvm_read_guest(kvm, gpa, data, len);
348 srcu_read_unlock(&kvm->srcu, srcu_idx);
353 #ifdef CONFIG_HARDEN_BRANCH_PREDICTOR
356 static inline void *kvm_get_hyp_vector(void)
358 struct bp_hardening_data *data = arm64_get_bp_hardening_data();
359 void *vect = kvm_ksym_ref(__kvm_hyp_vector);
362 vect = __bp_harden_hyp_vecs_start +
363 data->hyp_vectors_slot * SZ_2K;
366 vect = lm_alias(vect);
372 static inline int kvm_map_vectors(void)
374 return create_hyp_mappings(kvm_ksym_ref(__bp_harden_hyp_vecs_start),
375 kvm_ksym_ref(__bp_harden_hyp_vecs_end),
380 static inline void *kvm_get_hyp_vector(void)
382 return kvm_ksym_ref(__kvm_hyp_vector);
385 static inline int kvm_map_vectors(void)
391 #ifdef CONFIG_ARM64_SSBD
392 DECLARE_PER_CPU_READ_MOSTLY(u64, arm64_ssbd_callback_required);
394 static inline int hyp_map_aux_data(void)
398 for_each_possible_cpu(cpu) {
401 ptr = per_cpu_ptr(&arm64_ssbd_callback_required, cpu);
402 err = create_hyp_mappings(ptr, ptr + 1, PAGE_HYP);
409 static inline int hyp_map_aux_data(void)
415 #endif /* __ASSEMBLY__ */
416 #endif /* __ARM64_KVM_MMU_H__ */